Copy-number and gene dependency analysis reveals partial copy loss of wild-type SF3B1 as a novel cancer vulnerability

copy loss of wild-type SF3B1 as a novel cancer vulnerability

The MIT Faculty has made this article openly available.

Please share

how this access benefits you. Your story matters.

Citation

Paolella, Brenton R; Gibson, William J; Urbanski, Laura M; Alberta,

John A; Zack, Travis I; Bandopadhayay, Pratiti; Nichols, Caitlin A et

al. “Copy-Number and Gene Dependency Analysis Reveals Partial

Copy Loss of Wild-Type SF3B1 as a Novel Cancer Vulnerability.”

eLife 6 (February 2017): e23268 © 2017 Paolella et al

As Published

http://dx.doi.org/10.7554/eLife.23268

Publisher

eLife Sciences Publications, Ltd.

Version

Final published version

Citable link

http://hdl.handle.net/1721.1/109875

Terms of Use

Creative Commons Attribution 4.0 International License

*For correspondence: brenton_paolella@dfci.harvard. edu (BRP); robin_reed@hms. harvard.edu (RR); Rameen_Beroukhim@dfci.harvard. edu (RB)

†_{These authors contributed}

equally to this work Competing interests: The authors declare that no competing interests exist. Funding:See page 22 Received: 14 November 2016 Accepted: 06 February 2017 Published: 08 February 2017 Reviewing editor: Michael R Green, Howard Hughes Medical Institute, University of

Massachusetts Medical School, United States

Copyright Paolella et al. This article is distributed under the terms of theCreative Commons Attribution License,which permits unrestricted use and redistribution provided that the original author and source are credited.

Copy-number and gene dependency

analysis reveals partial copy loss of

wild-type SF3B1 as a novel cancer vulnerability

Brenton R Paolella

1,2,3

_*

†

_{, William J Gibson}

1,2,3†

, Laura M Urbanski

1,3

,

John A Alberta

1,4

_{, Travis I Zack}

1,2,3

_{, Pratiti Bandopadhayay}

1,2,3,5

_,

Caitlin A Nichols

1,2,3

_{, Pankaj K Agarwalla}

6

_{, Meredith S Brown}

1,3

_,

Rebecca Lamothe

1,3

, Yong Yu

7

, Peter S Choi

2,3

, Esther A Obeng

2,8

, Dirk Heckl

8

,

Guo Wei

2

_{, Belinda Wang}

2,3

_{, Aviad Tsherniak}

2

_{, Francisca Vazquez}

2

_,

Barbara A Weir

2

_{, David E Root}

2

_{, Glenn S Cowley}

2

_{, Sara J Buhrlage}

1

_,

Charles D Stiles

1,4

, Benjamin L Ebert

2,8

, William C Hahn

2,3,9

, Robin Reed

7

_*,

Rameen Beroukhim

1,2,3,9

_*

1

_{Department of Cancer Biology, Dana-Farber Cancer Institute and Harvard Medical}

School, Boston, United States;

2

_{Broad Institute of Massachusetts Institute of}

Technology and Harvard University, Cambridge, United States;

3

_{Department of}

Medical Oncology, Dana-Farber Cancer Institute and Harvard Medical School,

Boston, United States;

4

_{Department of Neurobiology, Harvard Medical School,}

Boston, United States;

5

_{Department of Pediatric Oncology, Dana-Farber Cancer}

Institute and Harvard Medical School, Boston, United States;

6

_{Department of}

Neurosurgery, Massachusetts General Hospital, Harvard Medical School, Boston,

United States;

7

_{Department of Cell Biology, Harvard Medical School, Boston,}

United States;

8

_{Division of Hematology, Department of Medicine, Brigham and}

Women’s Hospital, Harvard Medical School, Boston, United States;

9

_{Department of}

Medicine, Brigham and Women’s Hospital and Harvard Medical School, Boston,

United States

Abstract

Genomic instability is a hallmark of human cancer, and results in widespread somatic

copy number alterations. We used a genome-scale shRNA viability screen in human cancer cell lines to systematically identify genes that are essential in the context of particular copy-number

alterations (copy-number associated gene dependencies). The most enriched class of copy-number associated gene dependencies was CYCLOPS (Copy-number alterations Yielding Cancer Liabilities Owing to Partial losS) genes, and spliceosome components were the most prevalent. One of these, the pre-mRNA splicing factor SF3B1, is also frequently mutated in cancer. We validated SF3B1 as a CYCLOPS gene and found that human cancer cells harboring partial SF3B1 copy-loss lack a

reservoir of SF3b complex that protects cells with normal SF3B1 copy number from cell death upon partial SF3B1 suppression. These data provide a catalog of copy-number associated gene

dependencies and identify partial copy-loss of wild-type SF3B1 as a novel, non-driver cancer gene dependency.

DOI: 10.7554/eLife.23268.001

Introduction

Despite recent advances in cancer therapeutics, there remains a dearth of effective treatments. Therefore, expanding the number of candidate therapeutic targets in cancer is crucial. Cancer ‘driver

genes’, which undergo positive selection due to their effects on oncogenes or tumor suppressor genes, represent cancer vulnerabilities that are broadly considered as potential therapeutic targets (Cheung et al., 2011;Eifert and Powers, 2012;Wang et al., 2015). However, alterations of non-driver genes, which do not contribute to oncogenesis but are nevertheless observed, represent an emerging class of candidate therapeutic target that have yet to be fully explored.

During the course of tumorigenesis, most cancers undergo somatic copy number alterations (SCNAs) affecting large fractions of the genome (See Appendix Note) (Beroukhim et al., 2010). Yet most genes affected by SCNAs likely do not contribute to oncogenesis and are therefore over-whelmingly genetically altered non-driver genes. Recently, our laboratory and others have described potential new therapeutic targets that occur as a result of SCNAs affecting non-driver genes. For example, partial copy-loss of the proteasome subunit PSMC2, or RNA polymerase subunit POLR2A sensitized cancer cells to further suppression of those genes (Liu et al., 2015; Nijhawan et al., 2012). This ‘CYCLOPS’ (Copy-number alterations Yielding Cancer Liabilities Owing to Partial losS) phenotype suggests that many additional cancer vulnerabilities exist as a result of SCNAs that affect non-driver genes, although some CYCLOPS genes may function as driver genes when affected by other genetic alterations besides partial copy-loss. The frequency of these CYCLOPS gene depen-dencies and their general features are largely unknown.

These CYCLOPS genes tend also to be cell essential genes. While essential genes would be expected to be poor therapeutic targets because of their requirement for survival in all tissues, ther-apeutic windows can still exist (Muller et al., 2015). Identifying which essential genes may be consid-ered CYCLOPS genes, and the mechanisms underlying how normal cells tolerate partial loss of function, is necessary for developing approaches to target those therapeutic windows.

The spliceosome is one such essential protein complex that can be therapeutically targeted in cancer. Previous work suggested spliceosome components were enriched as candidate CYCLOPS

genes (Nijhawan et al., 2012). However, spliceosome CYCLOPS dependencies have yet to be

vali-dated and the molecular mechanisms for how spliceosome CYCLOPS dependencies arise remain unclear. Compounds have been discovered that inhibit pre-mRNA splicing, with reports of broad anti-neoplastic effects (Webb et al., 2013). Furthermore, cancers can harbor recurrent mutations in splicing factors (Dvinge et al., 2016), including gain-of-function mutations in SF3B1 (Ellis et al., 2012; Harbour et al., 2013; Imielinski et al., 2012; Papaemmanuil et al., 2011; Wang et al., 2011;Yoshida et al., 2011) that can sensitize cells to spliceosome modulatory drugs (Obeng et al.,

2016). In addition to SF3B1 mutations, other genomic alterations in SF3B1, including copy number

alterations, may also unveil novel cancer vulnerabilities. The extent to which SF3B1 and other splicing factors can be leveraged as therapeutic targets in cancer is not fully understood.

We therefore sought to systematically evaluate the prevalence of CYCLOPS dependencies rela-tive to other SCNA-associated gene dependencies in cancer. Here, we report that CYCLOPS depen-dencies are the most enriched class of copy-number associated gene dependency, even more frequent than amplification of oncogene-addicted driver gene. We find that CYCLOPS genes tend to be a subset of essential genes for which there is little feedback regulation in their expression when altered by SCNAs. We also find that more CYCLOPS gene dependencies are associated with spliceosome components than with any other gene family.

We find that wild-type SF3B1 is a non-driver CYCLOPS gene dependency and describe the mech-anism behind this dependency. Furthermore, the molecular mechmech-anism of the SF3B1 CYCLOPS dependency is distinct from SF3B1 dependencies targeted by current spliceosome inhibitors. We also identify the deubiquitinase inhibitor (DUBi’s), b-AP15, can reduce SF3B1 protein levels and tar-get the SF3B1 CYCLOPS dependency. Moreover, DUBi’s may represent a general therapeutic approach to target CYCLOPS gene vulnerabilities. The identification of SF3B1 as a CYCLOPS gene highlights a previously unrecognized cancer vulnerability and implicates non-driver alterations of wild-type SF3B1 as a potential therapeutic target present in 11% of all cancers.

Results

Most copy-number associated cancer dependencies result from

genomic loss

We interrogated copy-number associated vulnerabilities genome-wide across 179 cell lines by

inte-grating gene dependency data from Project Achilles (Cowley et al., 2014) with copy-number calls

for 23,124 genes (Barretina et al., 2012) (Figure 1A). The gene dependency data represented the effects on proliferation of 55,416 shRNAs targeting 11,589 unique genes, processed by the ATARiS

method to estimate effects of ‘on-target’ shRNAs (Shao et al., 2013), which yielded 8724 unique

gene-level dependency scores. For every pair of genes in the general analysis, we calculated Pearson correlations between the copy-number of the first gene and the dependency score of the second; yielding 201,733,776 parings in total (Figure 1A). We calculated p-values for each correlation and q-values to correct for multiple hypotheses (see Materials and methods). In many cases, a single gene dependency profile correlated with copy-number profiles of multiple genes from a single genomic region. We considered these to represent a single ‘independently significant’ interaction with the overall copy-number of that region.

In the general analysis, we identified 50 independently significant copy-number:gene-dependency interactions (q < 0.25;Supplementary file 1A). Approximately two-thirds (33/50) of these interac-tions involved genes on separate chromosomes (trans interacinterac-tions). Among the 33 trans interacinterac-tions, 21 reflected sensitization to suppression of a gene resulting from copy-loss of a different genomic region; the other 12 resulted from copy-gains. The trans gene dependencies identified were

enriched for members of macromolecular complexes (q = 2.510 4_{). In three cases, copy-loss of a}

gene (PPP2CB, FUBP2, or MAGOH) was associated with increased sensitivity to suppression of its paralog (PPP2CA, FUBP1, and MAGOHB, respectively). In contrast, all but one interaction (16/17) between genes on the same chromosome (cis interactions) involved increases in sensitivity to

sup-pression of a gene that had undergone copy-loss (CYCLOPS genes) (Nijhawan et al., 2012). As a

result, 74% of all 50 significant copy-number:gene-dependency interactions were associated with copy-loss rather than gain (p=1.510 4_).

Most cancers exhibit losses of candidate CYCLOPS genes

Although CYCLOPS interactions represent 0.004% (8,724/201,733,776) of all potential interactions in the general analysis, they constituted one-third of all significant interactions, making them the most enriched class of copy-number synthetic lethal interactions we identified (Figure 1B). Their preva-lence is partly the result of frequent genomic loss in cancer genomes. Specifically, across 10,570 can-cers spanning 31 cancer types profiled by The Cancer Genome Atlas (TCGA), 16% of the genome

undergoes loss in the average cancer (Figure 1—figure supplement 1A), mainly due to losses

encompassing chromosome arms or entire chromosome (Figure 1—figure supplement 1B). Indeed,

loss of a tumor suppressor often involves such arm-level losses (Figure 1—figure supplement 1C). The fraction of the genome lost per tumor ranged from an average of 1.3% in thyroid cancer to 34.4% in ovarian cancer (Figure 1C).

To enhance the identification of CYCLOPS genes, we performed a genome-wide analysis focused

on just CYCLOPS dependencies (Figure 1D; Materials and methods). The CYCLOPS analysis had

greater power than the general analysis described above because it focused on fewer hypotheses.

From the CYCLOPS analysis, we identified 124 candidate CYCLOPS genes (Supplementary file 1B),

including 87% of the candidate CYCLOPS genes identified in the general analysis

(Supplementary file 1A). Candidate CYCLOPS genes were distributed across all autosomes (Figure 1E) and were biased towards areas of frequent copy-loss (p<10 15_{). The bias toward more}

frequent copy-loss of candidate CYCLOPS genes may reflect greater statistical power in frequently deleted regions. We also examined the reproducibility of the CYCLOPS analysis using a separate shRNA viability screen across 77 cell lines with a separate analytical approach (Marcotte et al.,

2016) and found 49% (49/103) of the CYCLOPS genes analyzed by both datasets validated in the

Marcotte dataset (p<210 4, binomial proportion test).

Partial copy-loss of candidate CYCLOPS genes is frequent in cancer genomes. Among the 10,570 TCGA cancers with copy-number data, 71.6% harbored loss of at least one candidate CYCLOPS

0 5 10 15 20 Normalized variance in expression

across normal tissues

All genes Essential genes CYCLOPS genes

D

A

C

Copy-numbers _dependenciesGene

Copy-number of gene X r etf a ytil i b ai V n oi s s er p p u s Y e n e g

For all gene pairs

+

s er o c s e di w e m o n e G ) s e nil ll e c 9 7 1( 50 significant interactions

si

s

yl

a

n

a l

ar

e

n

e

G

B

-4 -2 0 2 Neutral Hemizygous loss Copy-number of gene N r etf a ytli b ai V n oi s s er p p u s N e n e g

si

s

yl

a

n

a

S

P

O

L

C

Y

C

124 significant interactions 8,724 genes N umber of C YCLOPS ge nes lo st pe r tumor 0 50 10 0

THCA KIRP COAD BRCA LUAD LUSC OV

Tumor type -200 -100 0 100 200 CYCLOPS log p-value gain loss Interaction type: Depletion Enrichment gain loss trans cis 0 20 40 60

Percent of genome lost

o

v

kich ucs lusc esca luad sarc chol blca _skcm brca hnsc read lihc _meso cesc stad acc paad gbm coad pcpg kirc lgg ucec prad uvm dlbc kirp laml thca

E

0 0.5 1.0 1.5 Genomic Position Delta Ata ris Score

Hemizygous deletion frequency

CYCLOPS gene _0.75

0.5

0.25

0

Frequency of partial loss

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16171819 2122

F

G

Figure 1. Analysis of copy-number-associated gene dependencies. (A) Schematic describing the general analysis of copy-number-associated gene dependencies. (B) Significance of enrichment for positive copy-number:gene dependency associations across trans and cis association classes, relative to the expected share assuming all possible interactions exhibited equal likelihood of positive association. (C) Percent of genome lost across 31 TCGA cancer types. Cancer types are indicated by TCGA abbreviations (see https://tcga-data.nci.nih.gov/datareports/codeTablesReport.htm, ‘Disease Study’ Figure 1 continued on next page

gene (Figure 1F). The average number of candidate CYCLOPS genes lost per tumor ranged from one in thyroid cancer to 47 in ovarian cancer.

Candidate CYCLOPS genes tend to be uniformly expressed members of

essential complexes

Of the 124 candidate CYCLOPS genes, 20 were members of the spliceosome and 11 were members of the proteasome, making these the only significantly enriched KEGG pathways among candidate

CYCLOPS genes (Supplementary file 1D). Candidate CYCLOPS genes are also enriched for

essen-tial genes. Genome-wide CRISPR viability data in human cancer cell lines identified 1580 core-essen-tial genes (Hart et al., 2015), including 58% (72/124) of candidate CYCLOPS genes (p<210 4, binomial proportion test).

Expression of candidate CYCLOPS genes is markedly uniform across normal tissues. Across RNA-sequencing data from 2342 samples comprising 42 tissue types (Mele´ et al., 2015), expression of candidate CYCLOPS genes varied significantly less than that of the average gene (p=1.810 15) and

trended towards greater uniformity than non-CYCLOPS essential genes (p=0.07;Figure 1G).

However, expression of candidate CYCLOPS genes is highly responsive to genomic loss in cancer. We integrated expression profiles for 16,867 genes with copy-number profiles across 1011 cell lines in the Cancer Cell Line Encyclopedia (CCLE) (Barretina et al., 2012) to determine the influence of copy-number on gene expression. Upon copy-loss, CYCLOPS gene expression decreased by 28% on

average, relative to an 18% decrease among non-CYCLOPS genes (p<10 4_{,Figure 1—figure}

sup-plement 1D–E). These data suggest that cancers with copy-loss of CYCLOPS genes are likely to express them at lower levels than normal tissues.

SF3B1 is a CYCLOPS gene

SF3B1 was among the most significant candidate genes in our CYCLOPS analysis

(Supplementary file 1B), although it was not detected in our general analysis (Supplementary file

1A). The SF3B1 protein is one of seven subunits (SF3B1–5, SF3B14 and PHF5A) of the SF3b

com-plex, which is a constituent of the essential U2 snRNP splicing factor (Wahl et al., 2009). Cells with SF3B1 copy-loss exhibited significantly reduced viability upon partial SF3B1 suppression relative to

cells without SF3B1 copy-loss (mean dependency scores of 1.14 and 0.01 respectively, p<10 5),

which suggests that partial suppression of SF3B1 can be tolerated in certain contexts even though it is an essential gene.

SF3B1 is partially lost in 11% of the 10,570 cancers from the TCGA PanCan dataset (see Materials and methods for definitions of copy number states). Across all cancers SF3B1 copy-loss was 5.4

times more common than SF3B1 mutations, which occur in ~2% of all cancers (Supplementary file

1C), and mutations and copy-loss were mutually exclusive (p=0.007). Losses were more frequent in

breast (20%), urothelial bladder (32%) and chromophobe kidney cancers (71%). Genomic deletions of SF3B1 typically affect most of the chromosome arm (81% of losses) and are never homozygous (0/

10,570 cancers), consistent with characterization of SF3B1 as an essential gene (An and Henion,

2012;Isono et al., 2005). In contrast, 85% of genes are homozygously deleted at least once among the same 10,570 cancers. Similarly, analysis of copy number alterations from 1042 cancer cell lines in the CCLE indicated 24% of cell lines harbor partial SF3B1 deletion, including 16/61 (26%) of breast cancer cell lines, but never homozygous loss (0/1042 cell lines).

Figure 1 continued

table). (D) Schematic describing the approach to identify CYCLOPS genes. (E) Relative strength (delta ATARiS score) of CYCLOPS genes (orange circles; left axis) and frequency of hemizygous deletion (solid black line; right axis) against genomic position (x-axis). (F) The number of CYCLOPS genes lost per tumor for various tumor types as in (C). Horizontal black lines represent medians per tumor type. (G) Distribution of variances in gene expression for different gene classes, normalized to expression level (see Materials and methods), in normal tissues. Whiskers represent min/max values and boxes represent upper and lower quartile ranges. Width of plots represents relative sample density.

DOI: 10.7554/eLife.23268.002

The following figure supplement is available for figure 1:

Figure supplement 1. Analysis of copy-number-associated gene dependencies. DOI: 10.7554/eLife.23268.003

We established the vulnerability of SF3B1 cells to SF3B1 suppression in both breast and hematopoietic lineages. We tested the proliferation of six breast lines after partial SF3B1 suppres-sion, including three lines with SF3B1 copy-loss (SF3B1loss_{) and three without either loss or gain of}

the gene (SF3B1neutral_{). Upon partial SF3B1 suppression, SF3B1}loss_{cells exhibited significant growth}

defects but SF3B1neutral_{cells or SF3B1}gain_{cells did not (Figure 2AandFigure 2—figure supplement} 1A). Growth defects observed in SF3B1loss _{cells included BT549, a near triploid cell line with two}

copies of SF3B1. Partial SF3B1 suppression also decreased the growth of ESS1, an endometrial cell

line harboring an SF3B1K666N_{mutation (Figure 2AandFigure 2—figure supplement 1A). However,}

complete SF3B1 suppression resulted in growth defects even in SF3B1neutral_{cells (Figure 2—figure} supplement 1B–C), consistent with previous work establishing SF3B1 as an essential gene.

The SF3B1 shRNAs used to unveil the CYCLOPS vulnerability targeted separate regions of the gene and resulted in partial SF3B1 knockdown across all eleven breast cell lines (Figure 2B).

How-ever, SF3B1loss cells had greater levels of knockdown due to lower SF3B1 expression at baseline,

which was confirmed at the protein level in four of these lines by immunoblot (Figure 2—figure sup-plement 1D).

We also generated similar results in isogenic SF3B1losscells derived from the SF3B1neutralcell line

Cal51. We generated SF3B1loss cells using two independent CRISPR-Cas9 mediated methods of

gene editing (see Materials and methods). The first line contained a frameshift mutation inactivating one SF3B1 allele (SF3B1Loss-Cal51-1). The second line had deletion of one copy of the SF3B1 locus, generated by co-expressing two sgRNAs: one upstream targeting a heterozygous SNP, and one

downstream of SF3B1 (SF3B1Loss-Cal51-2). In both cases CRISPR-mediated SF3B1 loss resulted in

decreased proliferation upon SF3B1 suppression relative to cells that were generated in parallel but did not produce inactivating alleles (SF3B1control-Cal51cells;Figure 2AandFigure 2—figure supple-ment 1A).

We confirmed the vulnerability of the SF3B1loss cells to SF3B1 partial suppression using a GFP-competition assay in which we compared the proliferation rate of uninfected cells co-cultured with cells infected with a vector that co-expressed GFP and an shRNA targeting either LacZ or SF3B1. The expression of LacZ or SF3B1 shRNAs did not result in significant changes in proliferation of

SF3B1neutral cells in seven cell lines, including the non-transformed mammary cell line, MCF10A

(Figure 2C and Figure 2—figure supplement 1E). However, SF3B1loss cells expressing SF3B1

shRNAs were not compatible with long-term culture (Figure 2CandFigure 2—figure supplement

1E).

The SF3B1 CYCLOPS vulnerability is not recapitulated by suppression of other SF3b complex sub-units, and copy number alterations of other SF3b complex genes do not confer susceptibility to SF3B1 suppression. We calculated the significance of associations between Achilles RNAi depen-dency data of six of the seven SF3b complex subunits (we lacked RNAi data for SF3B2) and copy

numbers of the genes encoding them (Figure 2—figure supplement 1F). We identified associations

between copy numbers of several SF3b subunits and sensitivity to suppressing that same subunit, consistent with our prior determination that multiple SF3b complex subunits are candidate

CYCLOPS genes (Supplementary file 1B). However, we observed no associations between

suscepti-bility to suppression of any of the SF3b complex genes and copy numbers of different SF3b subu-nits. Further, we confirmed one comparison where suppression of PHF5A, an SF3b complex gene, did not alter the growth of SF3B1losscells (Figure 2—figure supplement 1G–H).

Partial suppression of SF3B1 leads to both cell cycle arrest and apoptosis in SF3B1loss but not SF3B1neutral_{lines. We generated cultures containing a tetracycline inducible system expressing}

hair-pins targeting Luciferase or SF3B1 (TR-shSF3B1#3 and an additional hairpin, TR-shSF3B1#5,

Fig-ure 2—figFig-ure supplement 1A), enabling us to discriminate SF3B1 suppression from infection with shRNA vectors. Consistent with stable SF3B1 suppression, inducible SF3B1 suppression retards

SF3B1loss _{cell growth and does not affect SF3B1}neutral _{growth (Figure 2—figure supplement 1B)}

and reduces cell viability in SF3B1losscells but not in SF3B1neutralcells (Figure 2D and E). SF3B1loss cells had significantly increased proportions of cells in G2/M phase after SF3B1 suppression, which did not occur in SF3B1neutral _{cells (Figure 2F). Subsequent to G2/M arrest, SF3B1}loss _{cells further}

exhibited a significant induction in apoptosis as determined by increased number of AnnexinV/PI-positive cells that was not observed in SF3B1neutralcells (Figure 2G).

Expression of exogenous SF3B1 rescued the loss of viability in SF3B1loss _{cells, confirming the}

3 4 5 6 7 0.0

0.5 1.0 1.5

Days post doxycycline

R e la tiv e v ia b ili ty SF3B1neutral _-Dox SF3B1neutral _+Dox SF3B1loss _-Dox SF3B1loss _+Dox * ** 4 5 6 7 0.0 0.5 1.0 1.5

Days post infection

R e la tiv e v ia b ili ty shLacZ shSF3B1 shLacZ shSF3B1 shLacZ shSF3B1 HMEC MCF10A HCC1954_SF3B1loss SF3B1neutral * *** 0 5 10 15 20 25 0.0 0.5 1.0 1.5 2.0

Days post infection

R a tio s h S F 3 B 1 -G F P + :s h L a c Z -G F P + _SF3B1neutral SF3B1loss ** *** *** 0 5 7 9 0 50 100 150 200 250

Days post doxycycline

P e rc e n t c h a n g e in li g h t u n its TR shSF3B1#5 +LacZ TR shSF3B1#5 +SF3B1

B

D

C

F

E

G

H

I

T47D 1 3 5 7 0 50 100 150 200 250 shLacZ shSF3B1 #3 shSF3B1 #4 HCC1954 0 2 4 6 8 0 200 400 600 800 BT549 1 3 5 7 0 100 200 300 400 Cal 51 1 3 5 7 0 100 200 300 400 500 HMC 1-8 1 3 5 7 0 100 200 300 400 500 0 2 4 6 8 0 100 200 300 400 Hs578T SF3B1neutral _SF3B1loss 0 2 4 6 8 0 500 1000 1500 SF3B1Loss-Cal51-1

A

SF3B1control-Cal51-1 0 2 4 6 8 0 500 1000 1500

Percent change in light units

Days post infection

G0/G1 S G2/M 0 20 40 60 80 SF3B1neutral % o f c e lls G0/G1 S G2/M 0 20 40 60 80 SF3B1loss -Dox +Dox (shSF3B1) ** _** -Dox +Dox (shSF3B1)

Viable Apoptotic Dead

0 20 40 60 80 100 % o f c e lls

Viable Apoptotic Dead

0 20 40 60 80 100 *** *** * SF3B1neutral _SF3B1loss -Dox +Dox (shSF3B1) -Dox +Dox (shSF3B1) 0 2 4 6 8 10 0.0 0.5 1.0 1.5 2.0

Days post doxycycline

F o ld c h a n g e (% o f c e lls ) No SF3B1 knockdown Cal51 control Cal51 +SF3B1 HCC1954 control HCC1954 +SF3B1 0 2 4 6 8 10 0.0 0.5 1.0 1.5 2.0 _{with SF3B1 knockdown}

Days post doxycycline

F o ld c h a n g e (% o f c e lls ) Cal51 shSF3B1 Cal51 shSF3B1 +SF3B1 HCC1954 shSF3B1 HCC1954 shSF3B1 +SF3B1 shLacZ shSF3B1 #3 shSF3B1 #4 HM C 1-8 Cal -51 Hs5 78T MC F7 MC F10A_HMEC ESS1 HC C195 4 T47D_BT54 9 SKBR 3 0.0 0.5 1.0 1.5 m R N A F o ld C h a n g e

J

Figure 2. Characterization of SF3B1 as a CYCLOPS gene. (A) Growth of breast cancer cell lines expressing shLacZ (black) or shSF3B1 (red and orange), measured as changes in CellTiterGlo luminescence relative to one day post-infection. (B) Quantitative RT-PCR of SF3B1 expression from the indicated cell lines expressing shLacZ or shSF3B1 shRNAs normalized to the diploid SF3B1neutral_{cell line Cal51. (C) Ratio of cells expressing shSF3B1-GFP relative}

to uninfected controls, normalized to the ratio of cells expressing shLacZ-GFP relative to uninfected controls. Data represent averages from SF3B1neutral

which is resistant to shRNA suppression, fused to an IRES GFP sequence (SF3B1WT-IRES-GFP). When

placed in competition, cells infected or not infected with SF3B1WT_{-IRES-GFP maintained constant}

ratios over 10 days (Figure 2H), suggesting that short-term expression of SF3B1 does not alter cellu-lar fitness in either SF3B1neutralor SF3B1loss cells. Next, we concomitantly suppressed endogenous SF3B1 in all cells and expressed SF3B1WT-IRES-GFP in ~50% of cells. While SF3B1neutralcells were not affected by partial SF3B1 suppression, SF3B1loss_{cells expressing shSF3B1 were not compatible}

with long-term culture. However, SF3B1loss _{cells expressing both shSF3B1 and SF3B1}WT_-IRES-GFP

persisted (Figure 2I), indicating that re-expression of SF3B1 is sufficient to prevent cell death. Fur-thermore, SF3B1losscells expressing both shSF3B1 and SF3B1WT-IRES-GFP had a 20-fold increase in GFP fluorescence, suggesting that the exogenous SF3B1 construct was more highly expressed in SF3B1loss_{cells after suppression of endogenous SF3B1 (Figure 2—figure supplement 1C).}

Further-more, stable exogenous SF3B1 expression is sufficient to restore the proliferation of SF3B1losscells

expressing shRNAs targeting SF3B1 (Figure 2JandFigure 2—figure supplement 1D).

SF3B1

neutral

cells contain excess SF3B1 beyond the requirement for

survival

Analyses of SF3B1 mRNA indicate that SF3B1neutralcells tolerate partial SF3B1 suppression because they express more SF3B1 than the minimum amount needed for survival. In both TCGA breast

ade-nocarcinoma data (Cancer Genome Atlas Network, 2012) and the Cancer Cell Line Encyclopedia,

SF3B1neutralsamples exhibited significantly higher expression of SF3B1 mRNA relative to SF3B1loss

samples (Figure 3A andFigure 3—figure supplement 1A; Mann-Whitney p<10 4, for both

data-sets), suggesting excess mRNA over requirements for survival. We validated that SF3B1neutral_breast

cancer cell lines (n = 7) express approximately twice as much SF3B1 mRNA as SF3B1loss_{cells (n = 5)}

by quantitative PCR (Figure 3B; p<10 4). We also found similar SF3B1 expression changes between the SF3B1control-Cal51and SF3B1Loss-Cal51lines (Figure 3—figure supplement 1B).

Reductions in SF3B1 mRNA expression were recapitulated at the protein level. Among breast

cancer lines, we found increased SF3B1 protein expression in SF3B1neutral compared to SF3B1loss

cells (Figure 3C) and SF3B1control-Cal51 vs. SF3B1Loss-Cal51cells (Figure 3D). We also a found

signifi-cant linear correlation between SF3B1 mRNA and protein expression (Figure 3E, p=0.0018,

R2_{= 0.772).}

These observations suggest that SF3B1neutral _{cells tolerate partial SF3B1 suppression because}

moderate SF3B1 suppression leaves them with sufficient residual protein for survival. Indeed, we found detectable SF3B1 levels in SF3B1neutralcells after SF3B1 suppression, but failed to detect pro-tein in SF3B1losscells after SF3B1 suppression (Figure 3FandFigure 2—figure supplement 1D).

A systematic analysis of shRNA-induced mRNA suppression across SF3B1neutraland SF3B1losslines indicated that SF3B1 mRNA levels can be reduced by ~60% from SF3B1neutralcell basal levels before

proliferation and viability defects are apparent (Figure 3G). We suppressed SF3B1 using shRNAs

with different potency to generate a range of SF3B1 suppression in SF3B1neutral_{and SF3B1}loss_cells,

including shRNAs that suppress SF3B1 by >70% in SF3B1neutral_{cells, and determined the impact on}

Figure 2 continued

(n = 7) and SF3B1loss_{(n = 6) cell lines, using shSF3B1 #4 inFigure 2—figure supplement 2E. (D) Viability of cells expressing TR-shSF3B1#3 and}

TRshSF3B1#5, relative to viability three days post doxycycline administration. (E) Viability of cells expressing shLacZ or the average of shSF3B1#3 and shSF3B1#4, measured asthe fraction of propidium iodide negative cells, relative to the viability of these cells four days post infection. (F) Cell cycle distribution four days after SF3B1 suppression averaged from TR shSF3B1#3 and #5. (G) Fraction of apoptotic cells five days after SF3B1 suppression averaged from TR shSF3B1#3 and #5, as determined by AnnexinV/PI flow cytometry. (H) Change in ratio of cells expressing SF3B1-GFP relative to uninfected cells. (I) Ratio of cells expressing SF3B1-GFP to uninfected cells, in the context of endogenous SF3B1 suppression with TR-shSF3B1 #5. (J) Growth of LacZ and SF3B1 expressing SF3B1loss_{cells upon SF3B1 suppression (TR-shSF3B1#5), measured as changes in CellTiter-Glo luminescence. For}

all panels, *p<0.05 **p<0.01 ***p<0.001, and error bars represent ± SD from at least three (panels A–G) or two (H–J) replicates. DOI: 10.7554/eLife.23268.004

The following figure supplements are available for figure 2:

Figure supplement 1. Characterization of SF3B1 as a CYCLOPS gene. DOI: 10.7554/eLife.23268.005

Figure supplement 2. Further characterization of SF3B1 as a CYCLOPS gene. DOI: 10.7554/eLife.23268.006

A

C

E

F

Loss Neutral Gain

0 50 100 150 SF3B1 Copy Number SF3B1 expression (RPKM) *** *** 777 breast cancers

_B

SF3B1neutral _SF3B1loss 0.0 0.5 1.0 1.5 R e la tiv e S F 3 B 1 e x p re s s io n ***

Breast cancer cell lines

Actin

SF3B1

Cal51 Hs578 T MCF7 BT549 HCC1954ZR-75-30

SF3B1

neutral

_SF3B1

loss

D

0 .0 0 .5 1 .0 1 .5 R e la ti v e SF 3 B1 e x p re ssi o n 0.0 0.2 0.4 0.6 0.8 1.0 Proliferation SF3B1neutral _SF3B1loss 1 5l a C C E M H A 0 1 F C M 8-1 C M H 7 F C M T 8 7 5 s H 4 5 9 1 C C H D 7 4 T 9 4 5 T B Cell line:

SF3B1

Actin

Cal51

CRISPR line:

control-1

lo

ss-1

contr

ol-2

loss-2

G

0.0 0.5 1.0 0.0 0.5 1.0 Relative SF3B1 mRNA R e la ti v e S F 3 B 1 p ro te in R2 = 0.722 p = 0.018

Figure 3. SF3B1neutral_{cells contain excess SF3B1 beyond the requirement for survival. (A) SF3B1 expression from 777 TCGA breast adenocarcinomas}

segregated by SF3B1 copy number. Whiskers represent min/max values and boxes represent upper and lower quartile ranges. Width of plots represents relative sample density. (B) Quantitative RT-PCR of SF3B1 expression in breast cancer cell lines. Data points represent individual cell lines, horizontal lines indicate means. (C) SF3B1 protein levels in breast cancer cell lines by immunoblot. (D) SF3B1 immunoblot from control cells and those Figure 3 continued on next page

cellular growth seven days after shRNA expression. Although similar reductions in SF3B1 expression were obtained in SF3B1neutral_{and SF3B1}loss_{lines, the elevated basal levels of SF3B1 expression in}

SF3B1neutrallines enabled them to retain sufficient SF3B1 for proliferation despite shRNA expression, except in cases when SF3B1 suppression exceeded the 60% threshold of viability.

SF3B1 copy-loss selectively reduces the abundance of the SF3b

complex

We next asked whether the reduction of SF3B1 protein expression in SF3B1losscells preferentially altered specific SF3B1-containing complexes. SF3B1 is a component of the seven-member SF3b sub-complex of the U2 snRNP. Incorporation of SF3b into the U2 snRNP 12S ‘core’ forms the 15S U2 snRNP, which combines with SF3a to form the mature 17S U2 snRNP (Figure 4A) (Kra¨mer et al., 1999;van der Feltz et al., 2012). We therefore interrogated expression levels of native SF3B1-con-taining complexes from whole-cell extracts by glycerol gradient sedimentation and gel filtration chromatography. We were able to resolve protein complexes from 29–650 kDa and 650–2,000 kDa using 10–30% glycerol gradients and Sephacryl S-500 gel filtration chromatography, respectively (Figure 4—figure supplement 1A–B). This enabled resolution of SF3B1-containing complexes rang-ing from monomers (155 kDa) to the SF3b sub-complex (450 kDa) to the 15S and 17S U2 snRNPs (790 and 987 kDa, respectively) (van der Feltz et al., 2012). We compared these elution profiles between patient-derived and isogenic SF3B1lossand SF3B1neutralcells.

We observed significantly lower levels of SF3b in the SF3B1loss cells. The largest decreases in SF3B1-containing complexes in glycerol gradients were in fractions 4–8, corresponding to ~29–450

kDa (Figure 4B–F) and fractions 12–14, corresponding to ~450–650 kDa (Figure 4GandFigure 4—

figure supplement 1C). We saw similar decreases in gel filtration chromatography fractions

corre-sponding to complexes <650 kDa (Figure 4—figure supplement 1D). Native western blotting from

the pooled glycerol gradient fractions 4–6 indicated the loss of a single SF3B1-containing complex of ~450 kDa (Figure 4H). SF3B1 immunoprecipitation from fractions 4–6 resulted in the

co-precipita-tion of SF3b components SF3B3 and SF3B4 in SF3B1neutral_{cells, but not of U2 snRNP components}

SNRPB2 and SF3A3 (Figure 4—figure supplement 1E).

Conversely, it appears that U2 snRNP levels are only modestly decreased in SF3B1losslines. Levels of SF3B1 in glycerol gradient fraction 25 (corresponding to >650 kDa) were not significantly decreased in SF3B1loss relative to SF3B1neutral lines (Figure 4G andFigure 4—figure supplement 1C). SF3B1 immunoprecipitation from fractions 24–25 resulted in co-precipitation of U2 snRNP

com-ponents SNRPB2 and SF3A3 (Figure 4—figure supplement 1F). U2 snRNA levels are known to track

with U2 snRNP levels, and we also did not observe a significant difference in U2 snRNA abundance between SF3B1neutraland SF3B1losslines (Figure 4I p=0.35, two-tailed t-test). Similarly, visualization of U2 snRNP complexes using radiolabeled oligonucleotides complementary to the U2 snRNA did

not demonstrate differences in 17S U2 snRNP abundance in SF3B1loss _{cells (Figure 4J–K and}

Fig-ure 4—figFig-ure supplement 1G, p=0.68).

These observations suggest that copy-loss of SF3B1 only modestly affects U2 snRNP abundance but substantially decreases levels of U2 snRNP precursor complexes under steady-state conditions (Figure 4L).

Figure 3 continued

with single-copy SF3B1 inactivation by CRISPR. (E) Scatterplot of SF3B1 mRNA and protein expression relative to diploid cell line Cal51 after

normalization to actin in a panel of breast cancer cell lines (p=0.0018, R2= 0.772, regression line slope = 0.789). (F) SF3B1 immunoblot from SF3B1neutral and SF3B1loss_{cells 4 days after TR-shSF3B1#5 induction by doxycycline. (G) Differences in proliferation 7 days after SF3B1 suppression (per}

CellTiter-Glo, see Appendix Methods; red=high, blue=low), against the relative level of SF3B1 expression (assessed by qPCR; y-axis) in SF3B1neutral(left) or SF3B1loss_{(right) cells expressing either shLacZ (origins of arrows) or shSF3B1 (ends of arrows). Origins with multiple arrows represent cell lines subject to}

more than one SF3B1 shRNA. Each data point represents the mean from at least two replicate experiments. The dashed line represents the estimated minimum threshold of SF3B1 expression required for survival. For all panels, *p<0.05 **p<0.01 ***p<0.001, and error bars represent ± SD.

DOI: 10.7554/eLife.23268.007

The following figure supplement is available for figure 3:

Figure supplement 1. Further characterization of SF3B1 as a CYCLOPS gene. DOI: 10.7554/eLife.23268.008

C

Actin Cal 51 Hs578TBT549 HCC1954 SF3B1loss SF3B1neutral Gradient Input

SF3B1

8 7 6 5 4 3 3 4 5 6 7 8 3 4 5 6 7 8 3 4 5 6 7 8 Fraction #: Cal51 Hs578T HCC1954 BT549

SF3B1

neutral

_SF3B1

loss

E

SF3B1Neutral _SF3B1Loss HS578T CAL51 HCC1954 BT549 SF3B1 Fractions 4-6

Native Western Denaturing silver stain (loading control) SF3B1Neutral SF3B1Loss HS578 T CAL51 HCC195 4 BT5 49 Fractions 4-6 650 400

B

uL of sample: Fraction #: 4-6 12-14 25 Cal51 Hs578T BT549 HCC1954 20 10 5 2.5 20 10 5 2.5 20 10 5 2.5 20 10 5 2.5

SF3B1

neutral

_SF3B1

loss

}

1B 3 F S

F

A

D

SF3B1 SF3b 12S U2 snRNP 15S U2 snRNP SF3a 17S U2 snRNP SNRPB2 SF3A3 Protein Complex: SF3b subunits

Actin

Cal51

CRISPR line: control-1loss-1

Gradient Input Fraction: 3 4 5 6 7 8 3 4 5 6 7 8 SF3B1

SF3B1

control-Cal51-1

_SF3B1

Loss-Cal51-1 SF3B1neutral _SF3B1loss 0.0 0.5 1.0 1.5 U 2 s n R N A f o ld c h a n g e (r e la ti v e t o 5 .8 S ) ns

G

H

I

SF3B1loss 17S U2 snRNP SF3B1 SF3b 17S U2 snRNP SF3B1neutral

Model for U2 snRNP assembly

SF3B1 SF3b 3 4 5 6 7 8 0 50 100 150 200 200 400 600 800 1000 Fraction F o ld c h a n g e i n w e s te rn s ig n a l re la ti v e t o f ra c ti o n 3 SF3B1 neutral SF3B1loss * * *

J

K

L

SF3B1control-Cal51 _SF3B1Loss-Cal51 0.0 0.5 1.0 1.5 2.0 R e la ti v e 1 7 S U 2 s n R N P _ns - + + control-1Loss-1cont rol-2 Loss -2 + + 12S 17S ATP: + HeLa NE Cal51 CRISPR

Figure 4. SF3B1 copy-loss selectively reduces the abundance of the SF3b complex. (A) Diagram of U2 snRNP assembly. (B–C) Glycerol gradient fractionation of native whole-cell lysates probed by western blot in breast cancer cell lines and (D–E) isogenic cells generated by CRISPR. (F) Quantification of SF3B1 immunoblots from glycerol gradient fractions 3–8, relative to fraction 3 (n = 3 for each group, see Appendix Methods). (G) Serial dilution of pooled glycerol gradient fractions probed for SF3B1 by immunoblot. (H) (left) SF3B1 Native PAGE immunoblot of pooled glycerol Figure 4 continued on next page

SF3B1 suppression selectively reduces U2 snRNP abundance in

SF3B1

loss

cells

Partial suppression of SF3B1 leads to substantial reductions of U2 snRNP levels in SF3B1lossbut not SF3B1neutralcells. Although such suppression results in reduced SF3B1 levels in both SF3B1loss and SF3B1neutral_{lines, only the SF3B1}loss_{lines exhibit concomitant reductions in levels of U2 snRNP}

com-ponents SF3A3 and SNRPB2 (Figure 5A). Decreases in SF3A3 and SNRPB2 were observed in

glyc-erol gradient fraction 25, corresponding to the U2 snRNP, most dramatically in SF3B1losslines, and to a lesser extent in one of the SF3B1neutrallines (Hs578T;Figure 5B). Furthermore, after SF3B1 sup-pression, we detected both SF3B1 and SNRPB2 in Sephacryl-S500 fractions containing the >650 kDa protein complexes in SF3B1neutral_{cells but not in SF3B1}loss_{cells (Figure 5D–EandFigure 5—figure} supplement 1A). Quantitative PCR also indicated significantly reduced U2 snRNA expression after SF3B1 partial suppression in SF3B1loss_{cells but not in SF3B1}neutral_{cells (Figure 5F).}

In contrast, suppression of SF3B1 in SF3B1neutral_{cells appears to substantially decrease levels of}

SF3b, but not the U2 snRNP. SF3B1 suppression in SF3B1neutral_{cells only modestly reduced SF3B1 in}

fraction 25 (Figure 5B and Figure 5—figure supplement 1B) but instead preferentially reduced

SF3B1 from fractions 4–6 (Figure 5CandFigure 5—figure supplement 1C). Further, no changes in

SF3A3 or SNRPB2 expression were observed in total protein from glycerol gradient inputs (Figure 5A) or U2 snRNA expression (Figure 5F). Therefore, we determined if SF3B1 suppression in SF3B1neutral_{cells phenocopies the reduced SF3b observed in unperturbed SF3B1}loss _{cells. Indeed,}

SF3B1neutral cells with SF3B1 suppression reduced SF3b levels in glycerol gradient fractions 3–6

approximately to the levels observed in SF3B1losscells (Figure 5G). Taken together, these data sug-gest that the elevated levels of the SF3b sub-complex in SF3B1neutralcells relative to SF3B1losscells appear to buffer SF3B1neutralcells from reductions in viability following partial SF3B1 suppression.

Consistent with Figure 4, SF3B1loss _{cells without SF3B1 suppression did not have significantly}

lower levels of U2 snRNP in fraction 25 than SF3B1neutral_{cells (p=0.16,Figure 5—figure supplement} 1D), suggesting SF3B1 copy-loss only reduces levels of the U2 snRNP following further depletion of SF3B1.

The relative preservation of the U2 snRNP and larger complexes in SF3B1losscells without SF3B1 suppression suggests existing SF3b inhibitors might not exploit the specific vulnerability exhibited by SF3B1losscells. SF3b inhibitors modulate U2 snRNP function or subsequent steps during splicing catalysis (Corrionero et al., 2011; Folco et al., 2011; Roybal and Jurica, 2010), thereby altering splicing leading to cell death. With similar U2 snRNP levels in SF3B1loss _{and SF3B1}neutral _{cells, the}

dose at which these effects would be expected to accrue might be similar.

We tested this hypothesis by exposing SF3B1neutral_{cell lines with partial SF3B1 suppression, and}

controls to treatment with two SF3b-targeting compounds (Spliceostatin A, and Pladienolide B) and with NSC95397, a compound reported to inhibit splicing activity by an SF3b-independent mecha-nism (Berg et al., 2012). None of these exhibited increased effects on cells with partial SF3B1

sup-pression. (Figure 5—figure supplement 1E–H). We also evaluated isogenic Cal51 cells with

engineered SF3B1 copy-loss and did not observe enhanced sensitivity of SF3B1loss-Cal51 _{cell lines}

(Figure 5—figure supplement 1I).

Figure 4 continued

gradient fractions. (right) denaturing silver stain of total protein from the same pooled fractions as loading control. (I) Quantitative RT-PCR for U2 snRNA expression in three SF3B1neutral_{and three SF3B1}loss_{breast cancer cell lines. (J) Native agarose gel of U2 snRNP complexes visualized with}

radiolabeled 2’ O-methyl oligonucleotides complementary to the U2 snRNA. Nuclear extracts were generated from SF3B1control-Cal51_{and SF3B1}Loss-Cal51

cells. HeLa cell nuclear extracts (NE) ± ATP were used as controls. Representative image from triplicate experiments. (K) Densitometric quantification of 17S U2 snRNP bands in (J) presented as fold change relative to SF3B1control-Cal51_{cells. Data are from three replicate experiments. (L) Model for changes}

to U2 snRNP assembly associated with SF3B1 copy-loss. For all panels, *p<0.05, and error bars represent ± SD, ns = not significant. DOI: 10.7554/eLife.23268.009

The following figure supplement is available for figure 4:

Figure supplement 1. SF3B1 copy-loss selectively reduces the abundance of the SF3b complex. DOI: 10.7554/eLife.23268.010

Partial SF3B1 suppression results in splicing defects in SF3B1

loss

cells

SF3B1 is well-established as a splicing factor (Gozani et al., 1998;Zhou et al., 2002), and intron

retention has been reported upon treatment of cells with SF3B1 inhibitors (Kotake et al., 2007)

while patients harboring SF3B1 mutations display alterations in alternative splicing (DeBoever et al., 2015; Wang et al., 2011). We therefore quantified the extent of splicing disruption upon SF3B1 suppression in SF3B1neutral_{and SF3B1}loss_{cells by RNA sequencing. RNA was isolated 4 days after}

doxycycline treatment when SF3B1loss_{cells arrest in G2/M phase of the cell cycle (Figure 2F), but}

have not initiated apoptosis (Figure 2G). We suppressed SF3B1 in SF3B1neutralcells to similar or lower levels as seen at steady state in SF3B1loss cells, and suppressed SF3B1 in SF3B1loss cells to

even lower levels (Figure 3F and Figure 6—figure supplement 1A–B). We used juncBase

(Brooks et al., 2011) and a novel statistical framework to analyze 50,600 splice junctions for intron retention in SF3B1neutral_{and SF3B1}loss_{cells upon SF3B1 suppression from our RNA sequencing data.} SF3B1 SF3A3 SNRPB2 Actin Cal51 Hs578T BT549 HCC1954 Dox: SF3B1neutral _SF3B1loss - + - + - + - + Gradient Inputs Column Input

A

B

C

D

E

F

Hs578T Cal51 BT549 HCC1954 Dox:

- + - + - + - +

SF3B1

SF3A3

SNRPB2

Fraction 25 SF3B1neutral _SF3B1loss Hs578T Cal51 BT549 HCC1954 Dox:

- + - + - + - +

SF3A3

SNRPB2

Pooled Fractions 4-6

SF3B1

SF3B1neutral _SF3B1loss Fraction: 3 Cal51 -dox Cal51 +dox Hs578T -dox Hs578T +dox HCC1954 -dox BT549 -dox SF3B1 4 5 6 3 4 5 6 3 4 5 6 3 4 5 6 3 4 5 6 3 4 5 6

SF3B1

neutral

_SF3B1

loss

G

HCC1954 +dox Cal51 +dox

SF3B1

SNRPB2

SF3B1

neutral

_SF3B1

loss 26 25 24 23 22 21 20 19 18 1819202122 23 242526 Fraction: Cal51 TRsh5 HCC1954 TRsh5 Dox:

-

+

-

+

SF3B1

Actin

Cal51 Hs578T BT549 HCC1954 0.0 0.5 1.0 1.5 U 2 s n R N A f o ld c h a n g e _-Dox +Dox ** *** SF3B1neutral _SF3B1loss

Figure 5. Reduced spliceosome precursors and U2 snRNP abundance upon SF3B1 suppression in SF3B1loss_{cells. (A) Western immunoblots without and}

with SF3B1 suppression prior to glycerol gradient fractionation. (B) Western immunoblots from glycerol gradient fraction 25 (protein complexes >650 kDa). (C) Western immunoblots from pooled glycerol gradient fractions 4–6 (protein complexes ~150–450 kDa). (D) Immunoblot from lysates prior to gel filtration chromatography. (E) Immunoblot of gel filtration fractions 18–26 (protein complexes >650 kDa) from lysates with SF3B1 suppression. (F) Quantitative RT-PCR for U2 snRNA expression without and with SF3B1 suppression. (G) Glycerol gradient fractions from SF3B1neutral_{cells without and}

with SF3B1 suppression compared to SF3B1losswithout suppression. For all panels, TR-shSF3B1#5 was used. *p<0.05 **p<0.01 ***p<0.001. DOI: 10.7554/eLife.23268.011

The following figure supplement is available for figure 5:

Figure supplement 1. Reduced spliceosome precursors and U2 snRNP abundance upon SF3B1 suppression in SF3B1loss_cells.

Briefly, we calculated the ratio of percent spliced in (PSI) and spliced out read counts for each splic-ing junction, but accounted for the probability that any ssplic-ingle splicsplic-ing junction may not be accurately sampled in each cell line using a beta binomial distribution (see Appendix Methods).

All cells showed evidence of increased intron retention following SF3B1 partial suppression (p<10 5), but splicing was significantly more affected in SF3B1loss cells. Upon SF3B1 suppression, 7353 transcripts in SF3B1loss_{cells showed evidence of significantly (q < 0.1) increased intron}

reten-tion relative to SF3B1neutralcells, whereas only 454 transcripts showed evidence of increased intron retention in the reverse direction (Figure 6A, p<10 1110_).

We confirmed increased intron retention in SF3B1loss _{cells upon SF3B1 suppression by qPCR}

amplifying selected introns of genes from the RNA sequencing analysis (Figure 6A). Introns were

significantly retained in all seven genes analyzed (AARS, CALR, DNAJB1, MKNK2, MYH9, RPS8 and RPS18), which included DNAJB1, a gene previously known to be improperly spliced in cells treated

with SF3B1 inhibitors (Kotake et al., 2007), and cell essential genes AARS, RPS8 and RPS18

(Figure 6B).

The SF3b complex is known to regulate 3’ splice site selection (DeBoever et al., 2015). We

therefore analyzed 30,666 alternative 3’ splice sites from the RNAseq data in SF3B1neutral and

SF3B1loss _{cells. Reduced SF3B1 expression resulted in significantly more alternative 3’ splice site}

selection in SF3B1loss cells (353 3’ splice junctions in SF3B1neutral vs. 1540 in SF3B1loss cells, p<10 121_{,Figure 6CandFigure 6—figure supplement 1C). We also observed other alterations in}

splicing. Alternative 5’ splice site selection occurred at a significantly higher rate upon SF3B1 sup-pression in SF3B1loss_{cells (1411 junctions vs 317, p=910} 165_{,Figure 6—figure supplement 1D).}

We also observed increased dysregulation of cassette exon inclusion. The proportion of reads including the cassette exon at each junction changed substantially after SF3B1 suppression in SF3B1loss_{cells, but was less substantially altered in SF3B1}neutral_{cells. (Figure 6—figure supplement} 1E). We further validated these observations with independent assays for alternative splicing. Specif-ically, the ratio between alternative long and short isoforms of MCL1 (that respectively do or do not have anti-apoptotic functions) is known to be regulated by SF3B1 (Moore et al., 2010). After SF3B1 suppression, this ratio was significantly biased towards the short isoform in SF3B1loss_{cells relative to}

SF3B1neutralcells (Figure 6D–E). Together, these data indicate that SF3B1 suppression more sub-stantially dysregulates splicing of the transcriptome of SF3B1loss_cells.

Spliceosome components, including SF3B1, are thought to assemble and function in sub-nuclear

compartments known as nuclear speckles (Spector and Lamond, 2011). Inhibition of splicing or

transcription has been shown to induce formation of enlarged ‘mega-speckles’ (Kotake et al., 2007;

Loyer et al., 1998). We therefore performed an unbiased quantification of the number and size of

SC-35+ _{speckles per nucleus using a custom image analysis pipeline with CellProfiler software}

(Kamentsky et al., 2011).

SF3B1neutralcells did not display changes in SC-35+ speckles after SF3B1 partial suppression, but SF3B1loss nuclei contained significantly fewer speckles and increased speckle area (Figure 6F–H). The presence of defective alternative splicing, intron retention and formation of mega-speckles uniquely in SF3B1losscells after partial SF3B1 suppression further supports the gross defects in splic-ing observed by RNA sequencsplic-ing.

Moreover, upon SF3B1 suppression, 513 genes were differentially expressed at an FDR < 10% in

SF3B1loss cells and only 306 genes were differentially expressed in SF3B1neutral cells

(Supplementary file 1E, p<10 4_{, Fischer’s exact test). Gene set enrichment analysis revealed 24}

KEGG pathways significantly enriched in SF3B1losscells and only nine pathways altered in SF3B1

neu-tral_{cells (Figure 6IandSupplementary file 1F). We also identified a statistically significant core set}

of 89 genes differentially expressed across all cell lines upon SF3B1 suppression regardless of SF3B1 copy number (p<10 4_).

Upon SF3B1 expression, 348 genes were differentially expressed in SF3B1loss_{cells and 393 genes}

were differentially expressed in SF3B1neutral _{cells, an insignificant difference (p=0.07,}

Supplementary file 1G). We observed more differentially expressed genes in SF3B1loss cells from SF3B1 suppression than from SF3B1 expression (p<0.0001), however, there was no difference

between the effects of SF3B1 suppression and expression in SF3B1neutral _{lines (p=0.7). Taken}

together, these data indicate that partial SF3B1 suppression more severely impacts the transcrip-tome of SF3B1loss_cells.

B

A

SF3B1

neutral

_SF3B1

loss * shLacZ shSF3B1 0 2 4 6 8 F o ld C h a n g e (M C L 1 -s :M C L 1 -l r a ti o )

D

_E

F

_G

C

- Dox

+ Dox

Cal51

Hs578T

BT549

HCC1954

SF3B1

neutral

_SF3B1

loss

MCL1

Actin

Cal51 HCC 1954 c sh c sh SF3B1neutral_SF3B1loss MCL1-L MCL1-S

Alternative Splicing

Cal51 Hs578T BT549 HCC1954 0 50 100 150 200 S p e c k le a re a (s q u a re p ix e ls ) -Dox +Dox *** ***

SF3B1

neutral

_SF3B1

loss

Intron Retention

AAR S Cal R DNAJB1MK NK_M2YH_R9PS8 RPS 18 AAR S Cal R DNAJB1MK NK_M2YH_R9PS8 RPS 18 0 1 2 3 5 10 15 50 100 F o ld i n c re a s e in i n tr o n r e te n ti o n - Dox + Dox * *** *** *** *** *** *** * SF3B1neutral _SF3B1loss −4 −2 0 2 4

probability of alternative 3’ splicing (log ratio) more in SF3B1loss more in SF3B1neutral q-value 1 10-2 10-4 10-6 10-8 10-10 10-12 Cal51 Hs578T BT549 HCC1954 0 10 20 30 40 50 # S p e c k le s p e r c e ll *** ***

SF3B1

neutral

_SF3B1

loss -Dox +Dox

H

I

−4 −2 0 2 4

probability of intron retention (log ratio) more in SF3B1loss more in SF3B1neutral q-value 1 10-2 10-4 10-6 10-8 10-10 10-12 513 genes 306 genes 89 genes

SF3B1

loss

SF3B1

neutral

Differentially expressed genes/pathways

9 KEGG pathways

24 KEGG pathways

8 KEGG pathways

p<0.0001

Figure 6. Partial SF3B1 suppression results in splicing defects in SF3B1loss_{cells. (A) Statistical significance of intron retention (see}

Materials and methods) across all exon-intron junctions (dots) in SF3B1neutral_{(red) and SF3B1}loss_{cells (blue) after SF3B1 suppression. The horizontal}

dashed line represents the significance threshold (q < 0.01) and the vertical dashed line segregates intron-exon junctions more likely to be altered in SF3B1neutral(left) or SF3B1loss_{cells (right). (B) qPCR for a single intron within the indicated gene without and with shSF3B1 induction by doxycycline}

Suppression of SF3B1 reduces tumor growth in SF3B1

loss

xenografts

To test the effects of SF3B1 partial suppression in vivo, we generated xenografts from luciferase-labeled SF3B1Loss-Cal51-1_{, SF3B1}control-Cal51-1 _{cells and naturally occurring SF3B1}neutral _{and SF3B1}loss

cells (Cal51 and HCC1954, respectively) all containing TR-shSF3B1#3. Interventional studies were performed where animals were placed on doxycycline upon detection of palpable tumors and vali-dated for SF3B1 suppression (Figure 7—figure supplement 1A). SF3B1Loss-Cal51-1_{and SF3B1} control-Cal51-1 _{cells generated tumors of similar volume in the absence of doxycycline (Figure 7A; p=0.7,}

repeated measures ANOVA). However, partial suppression of SF3B1 reduced the growth of (Figure 7B–C) and number of proliferative Ki67+ cells in (Figure 7D–E) xenografts from SF3B1

Loss-Cal51-1

cells but not SF3B1control-Cal51-1 cells (p=0.001 for both assays). Similarly, reduced tumor

growth was observed in naturally occurring SF3B1loss HCC1954 xenografts and not in SF3B1neutral

Cal51 xenografts (Figure 7F).

Compounds that induce degradation of SF3B1 selectively kill SF3B1

loss

cells

The lack of efficacy of small molecule splicing modulators targeting SF3B1 prompted us to evaluate other compounds that could pharmacologically target the SF3B1 CYCLOPS dependency. Deubiquiti-nase inhibitors (DUBi’s) are a class of compounds that increase protein degradation by preventing the removal of ubiquitin from substrates, thereby enhancing target protein degradation by the pro-teasome. We first sought evidence for SF3B1 ubiquitination by treating cells with the proteasome inhibitor, epoxomicin, to increase the accumulation of ubiquitinated proteins. Immunoprecipitation of SF3B1 in denaturing conditions after epoxomicin treatment resulted in the detection of

ubiquiti-nated SF3B1 (Figure 8—figure supplement 1A–B). Furthermore, proteasome inhibition by MG-132

increased SF3B1 protein levels, suggesting SF3B1 is regulated by the proteasome (Figure 8—figure

supplement 1C).

To determine if DUBi’s could decrease SF3B1 protein expression, we evaluated seven candidate DUBi’s for reduced SF3B1 expression by western blot (data not shown) and identified three DUBi’s with the ability to reduce SF3B1 protein levels within 24 hr (Figure 8A). Of the 3 DUBi’s that enhance

SF3B1 degradation, one semi-selective DUBi b-AP15 (D’Arcy et al., 2015), demonstrated enhanced

sensitivity in SF3B1losscells, and also cells with reduced SF3B1 expression by RNAi (Figure 8B). The

enhanced sensitivity in SF3B1loss cells or cells with reduced SF3B1 expression by RNAi was not

observed upon treatment with SJB3-019A or the pan-DUBi, PR-619 (Figure 8—figure supplement

1D), suggesting inhibition of a specific DUB enzyme may mediate this effect. SF3B1Loss-Cal51 _cells

also had more Annexin V apoptotic cells than SF3B1control-Cal51 _{cells when treated with nanomolar}

concentrations of b-AP15 (Figure 8C). Immunoblots for SF3B1 after b-AP15 treatment revealed

decreased SF3B1 only occurred in SF3B1Loss-Cal51-2 _{cells (Figure 8D andFigure 8—figure}

supple-ment 1E). These data suggest the enhanced sensitivity may be due, in part, to reduction of SF3B1 in SF3B1Loss-Cal51cells.

Treatment of cells with b-AP15 resulted in splicing alterations preferentially in SF3B1loss_{cells. We}

determined the effect of b-AP15 treatment on previously identified splicing alterations that result

Figure 6 continued

(SF3B1neutral_{n = 3, SF3B1}loss_{n = 3, averaged from TR-shSF3B1#3 and TRshSF3B1#5. SF3B1}control-Cal51_{and SF3B1}Loss-Cal51_{n = 2 each, averaged from}

TR-shSF3B1#3). (C) Statistical significance of alternative 3’ splice site selection (see Materials and methods) across 3’ splice junctions (dots) in

SF3B1neutral_{(red) and SF3B1}loss_{cells (blue) after SF3B1 suppression (as in panel A). (D) Representative RT-PCR from SF3B1}neutral_{and SF3B1}loss_{cells after}

SF3B1 knockdown. ‘c’ represents LacZ control hairpins, ‘sh’ represents shSF3B1#4. Arrows represent product sizes for MCL-L and MCL-S. (E) Densitometric quantification of the ratio of MCL1-S:MCL1-L, relative to shLacZ-expressing controls (mean ± SD, n = 3 per group averaged from shSF3B1#3, and #4). (F) Immunofluorescent images of nuclear speckles by anti-SC35 (SRSF2) staining. Scale bar = 5 uM. (G) Quantification of number of nuclear speckles and (H) speckle area per cell across at least 100 nuclei. For (F–H) TRshSF3B1#5 was used. (I) Number of differentially expressed genes upon SF3B1 suppression (q < 0.1) and the number of enriched KEGG pathways amongst indicated gene set (q < 0.05). For all panels, *p<0.05 **p<0.01 ***p<0.001.

DOI: 10.7554/eLife.23268.013

The following figure supplement is available for figure 6:

Figure supplement 1. Partial SF3B1 suppression results in splicing defects in SF3B1losscells. DOI: 10.7554/eLife.23268.014

0 10 20 30 40 0.0 5.0 106 1.0 107 1.5 107 L u m e n e s c e n c e (p h o to n s /s e c ) Plus Dox

A

B

C

***

low high st n u o c e c n e c s e n e m ul

No dox Plus dox

Left flank:

SF3B1

control-Cal51-1

Right flank:

SF3B1

Loss-Cal51-1

0 10 20 30 40

0

1 107

2 107

3 107

Days post tumor detection

L u m e n e s c e n c e (p h o to n s /s e c ) No dox

D

E

SF3B1

control-Cal51-1

_SF3B1

Loss-Cal51-1

- Dox

+ Dox

No Dox Plus Dox

0.0 0.2 0.4 0.6 0.8

R

a

tio

K

i6

7

+

*** SF3B1control-Cal51-1 SF3B1Loss-Cal51-1 Dox added

Days post tumor detection

SF3B1control-Cal51-1 _SF3B1Loss-Cal51-1 10 20 30 40 0 500 1000 1500

Days post injection

Cal51 (SF3B1neutral₎ Dox added m m( e m ul o v r o m u T 3) 10 20 30 40 0 500 1000 1500

Days post injection

Dox added m m( e m ul o v r o m u T 3) HCC1954 (SF3B1 loss₎

***

F

-Dox

+Dox

Figure 7. SF3B1 suppression inhibits tumor growth in vivo. Luminescent quantification of xenograft growth from SF3B1control-Cal51-1_{(black) and}

SF3B1Loss-Cal51-1_{(red) tumors (A) without doxycycline (n = 4) and (B) with doxycycline (n = 17) using TR-shSF3B1 #3. (C) Representative animal images}

overlaid with heat maps from bioluminescent tumor detection. Dashed circle represents region where established tumor was detected prior to doxycycline treatment. (D) Quantification of Ki67+ cells from xenografts 42 days post tumor detection using CellProfiler  2440 nuclei were scored for each tumor,  3 tumors per group (see Materials and methods). (E) Representative Ki67 immunohistochemistry images of xenografts quantified in panel (D). (F) Growth of established tumors for Cal51 and HCC1954 xenografts without doxycycline (black) or with doxycycline (red) using TR-shSF3B1 #3. For Cal51 and HCC1954 no doxycycline groups (n = 13); plus doxycycline groups (n = 12). For all panels, ***p0.001.

DOI: 10.7554/eLife.23268.015

The following figure supplement is available for figure 7:

Figure supplement 1. SF3B1 suppression inhibits tumor growth in vivo. DOI: 10.7554/eLife.23268.016